A multi-objective macro-geometry spur gear optimization process to improve weight, efficiency and NVH performance utilizing neural networks

In this study, a macro-geometry optimization process is developed to improve the weight, efficiency, and Noise, Vibration, and Harshness (NVH) performance of spur gears. NVH improvement is achieved through the minimization of the peak-to-peak static transmission error (STE). STE curves are calculated using a fully connected feed-forward neural network (FFNN) that achieves a mean absolute percentage error below 0.5%. Based on the FFNN, a multi-objective optimization framework using the NSGA-II algorithm is developed and applied to seven case studies. The results demonstrate that macro-geometry optimization can significantly enhance NVH performance, achieving reductions in the RMS of dynamic transmission error exceeding 35%, while simultaneously reducing power losses by more than 40%. The findings reveal a strong correlation between power loss reduction and peak-to-peak STE and confirm peak-to-peak STE as an effective surrogate objective for improving dynamic behavior. The proposed framework enables rapid and accurate NVH-driven gear design by integrating neural-network-based STE prediction with multi-objective evolutionary optimization, demonstrating that substantial NVH improvements can be achieved through macro-geometry optimization alone. In addition, general guidelines for the selection of macro-geometric parameters are proposed.